OpenAI protocol endpoints

The proxy also implements the OpenAI Chat Completions API at /v1/*. This is what GitHub Copilot, the OpenAI Python SDK, Continue, Cursor, and many other tools speak.

Endpoint	Notes
POST /v1/chat/completions	Streaming SSE + non-streaming, tools, response_format
GET /v1/models	Lists registered models in OpenAI format

POST /v1/chat/completions

curl http://localhost:11434/v1/chat/completions -d '{
  "model": "gemini/gemini-2.5-flash",
  "messages": [{"role": "user", "content": "Hello!"}],
  "stream": false,
  "tools": [...],
  "response_format": {"type": "json_object"}
}'

Non-streaming response (truncated):

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "created": 1780488000,
  "model": "gemini/gemini-2.5-flash",
  "choices": [{
    "index": 0,
    "message": {"role": "assistant", "content": "Hello!"},
    "finish_reason": "stop"
  }],
  "usage": {"prompt_tokens": 5, "completion_tokens": 1, "total_tokens": 6}
}

Streaming

When stream: true, responses are Server-Sent Events:

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"!"},"finish_reason":null}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]

tool_calls deltas are streamed in the same format the OpenAI SDK expects, so Copilot and similar tools can drive function-calling flows correctly.

Parameter handling

OpenAI param	Behavior
model	<prefix>/<model-id> — required
messages	Standard OpenAI format, passed through
tools	Passed through to providers that support it
response_format	json_object and json_schema passed through
max_tokens / max_completion_tokens	Mapped to provider's max-tokens param
stop	Passed through
temperature / top_p	Skipped — see note below

Why temperature isn't forwarded

Some models (notably Kimi K2.5+) reject any temperature value other than 1. Copilot and other clients tend to set a default temperature like 0.7, which would cause hard failures for those models.

The proxy intentionally drops temperature and top_p from the /v1/chat/completions endpoint to avoid this. If you need custom temperature, use the Ollama options object on /api/chat instead.

Provider-specific behavior

Kimi

The proxy automatically sets thinking: {type: "disabled"} for kimi/ models on this endpoint. Copilot and similar clients can't render reasoning tokens, so disabling them keeps responses clean.

GET /v1/models

Returns all registered models in OpenAI list format:

{
  "object": "list",
  "data": [
    {"id": "openai/gpt-4o", "object": "model", "owned_by": "openai", "created": 1700000000},
    {"id": "gemini/gemini-2.5-pro", "object": "model", "owned_by": "gemini", "created": 1700000000}
  ]
}

See also

• Logging & tracing — every response has an x-request-id header you can grep for in logs.
• Retry & fallback — applies to this endpoint just like Ollama protocol calls.